GLACE Alps

Sentinel-1 interferometric coherence and backscatter composites of the European Alps

🗺️ Web map · 🧭 Portolan/STAC browser · 🏷️ Metadata repository · ⚙️ Production pipeline

Annual summer composites of Sentinel-1 12-day repeat-pass interferometric coherence (COH12) and radiometrically terrain-corrected backscatter (RTC) for the Alpine Convention perimeter, 2015–2026. The data is available as Cloud-Optimized GeoTIFFs (COGs), described by a STAC catalog.

Collections

Both represent the same source composites on different grids. The tiles hold the main product for scientific analysis. The mosaics put every year on one fixed equal-area grid, so years can be stacked and compared without alignment. The mosaics collection also includes PMTiles archives for web maps.

To explore the data visually, open the GLACE Viewer.

Get started

All files are public on Source Cooperative and can be read anonymously over HTTPS (https://data.source.coop/giuz/glace-alps/) or S3 (s3://eu-central-1.opendata.source.coop/giuz/glace-alps/). Each collection provides a searchable item index, items.parquet, in STAC-GeoParquet format: one file to search across all items.

This example uses rustac to find tiles for an area and year, then rasterio to read coherence values from the first matching tile. Install both packages before running it.

The bounding box selects whole tiles, so this reads the full first tile.

For more workflows, see building a data cube, opening tiles in QGIS, and working with mosaics.

Products

Each tile or annual mosaic contains twelve analysis rasters: coherence and backscatter in VV and VH polarization, plus an observation count and a composite quality map for each measurement. See the asset table for filenames, data types and nodata values.

Use the COGs for analysis. The _viz.pmtiles archives are for web maps: they hold lossy 8-bit value codes over a fixed range per layer, which the map colors in the browser. In the archives, backscatter and composite quality maps are in decibels and coherence is linear.

Every analysis raster carries its description and processing record in its GeoTIFF tags.

Usage notes

Composite quality. Lower *_qa_cqm values indicate a lower local resolution of the composite: the closer to 0, the fewer favourable viewing geometries contributed. In the validation, a random forest maps glaciers less accurately where the CQM is below 0.5 (linear, about −3 dB; AUC 0.87) than elsewhere (0.97). These pixels make up 3% of the pixels within 250 m of the glacier outlines and 1.5% of the glacier pixels. Mask or downweight them, or use the CQM as an additional model input.

Coherence bias. Coherence is estimated over a 35 × 7 (range × azimuth) window. The finite window biases the estimate upwards, so fully decorrelated surfaces do not reach 0: over large lakes, the mean is about 0.09. Absolute values therefore differ from products that use larger windows, such as the global seasonal Sentinel-1 coherence of Kellndorfer et al. (2022). Do not compare values across products, and choose thresholds on GLACE itself.

Low coherence outside glaciers. Other surfaces also decorrelate over 12 days and can have similarly low coherence, such as snow patches, lakes and unstable slopes.

Validated scope. The validation covers glacier mapping in Switzerland only. Its scores do not describe glacier mapping performance in other parts of the Alps, or farther than 250 m from the glacier outlines. Other applications may also benefit from the composites, but were not validated.

2015 coverage. In 2015, coherence is missing on 91 MGRS squares in the north-central Alps (100 km squares 32TNT, 32TNU, 32TPT, 32TPU and 32UPU) because Sentinel-1 coverage was limited that summer. Two of them, 32TNT14 and 32TNT24, have no 2015 items. Every other year covers all squares.

Composite processing

Sentinel-1 SLC bursts are processed to 12-day repeat-pass coherence and terrain-corrected gamma-nought backscatter, then composited over each year's 90-day summer window with local resolution weighting after Small et al. (2022). Local resolution depends on terrain slope and aspect relative to the radar viewing geometry. At each pixel, observations with finer resolution in map coordinates receive greater weight. Each year's window is chosen by minimizing MODIS NDSI over the target area.

See the production pipeline for processing details.

License and citation

CC-BY-4.0. Contains modified Copernicus Sentinel data.

Cite this version of the dataset as:

Gentner, L. Q. (2026). GLACE Alps (Version 1.0) [Data set]. LaRS – Language Repository of Switzerland. https://doi.org/10.48656/60er-3j55

To refer to the dataset across versions, use https://doi.org/10.48656/d4r9-fj97.

Feedback and contributions

Report problems or suggest improvements through the issue tracker. The catalog's metadata is maintained in the metadata repository.

For programmatic agent guidance, see AGENTS.md.